Study On Stock Market Prediction Based On Support Vector Machine

Financial stock market is not only a barometer in national economy, but also one of the most important investment tools and financingmeans for both personal and enterprices. If we can make some prediction and analysis before time, then take some action to cope with the financial crisis, this would make significantly theory contribution and real meaning for no matter who like private investor,enterprices and the government policy maker. Support Vector Machine is a new technology of data mining based on VC Dimension theory and Structural Risk Minimization(SRM) of Statistical Learning Theory(SLT) which has shown its great special advantages in hign dimension pattern recognition,function fitting and time series forecasting. This paper is organized as follows:In Section 1, a introduction about stock financial forecasting had been made after read lots of papers at home and abroad, and introduce some foundation knowledge of SLT and SVM theory in order to making the preparation for the next sections' works. In Section 2, in terms of the characteristic of financial data, a new kernel named financial kernel(f-kernel) had been constructed for the improvement of the prediction accuracy, the relevance parameters selection and control are also investigated. In Section 3, based on the f-kernel,several kinds of transformed support vector machines had been investigated and compared with normal SVR. In Section 4,according to the two key problem:noisy and non-stationary, the two-stage model based on k - means cluster method has been proposed, the shanghai stock composite index, wanke share and the fund jintai are used in this experiments. It is shown that the proposed model performed significantly higher prediction accuracy than the traditional single SVR model. Finally, fifty days roll prediction based on three typical stock market data had been given out, experiments show that this proposed method achieves good performance than triditional single SVM model both on learning capacity and generalization.